Upload
others
View
14
Download
0
Embed Size (px)
Citation preview
HRS/AHEAD Documentation Report
Documentation of Physical Functioning Measured in the Heath and Retirement
Study and the Asset and Health Dynamics among the Oldest Old Study
Report Prepared by Stephanie Fonda and A. Regula Herzog
With Contributions from Past and Present Members of the HRS Health Working Group:
Robert B. Wallace A. Regula Herzog
Mary Beth Ofstedal Stephanie J. Fonda
Diane Steffick Kenneth Langa
Gwenith G. Fisher Nancy Fultz David Weir Jessica Faul
Survey Research Center University of Michigan
Ann Arbor, MI
LAST UPDATED December 21, 2004
Table of Contents
Table of Contents ................................................................................................................................... ii I. Overview............................................................................................................................................. 1 II. Rationale for measuring physical functioning................................................................................... 2 III. Inventory of measures ...................................................................................................................... 5
A. Mobility, strength and fine motor skills......................................................................................... 5 B. Activities of daily living (ADLs) ................................................................................................... 7 C. Instrumental activities of daily living (IADLs)............................................................................ 11 D. Accommodations ......................................................................................................................... 13
IV. Pedigree of physical functioning measures.................................................................................... 14 V. Special methodological issues......................................................................................................... 16
A. Measurement error ....................................................................................................................... 16 B. Comparison of questions across waves........................................................................................ 17
VI. Evaluation of the measures and data quality.................................................................................. 23 A. Previous research on internal consistency and measurement properties ..................................... 23 B. A basic examination of internal consistency and measurement properties.................................. 25 C. Univariate distributions................................................................................................................ 28 D. Previous research on construct validity ....................................................................................... 36 F. Benchmarking against other surveys and prevalence estimates................................................... 37 G. Constructed variables................................................................................................................... 40
VII. Technicalities................................................................................................................................ 41 A. Strategies for accommodating changes in questions ................................................................... 41 B. SAS code...................................................................................................................................... 44
References ............................................................................................................................................ 57
ii
I. Overview This document is part of a series of working papers on health-related measures in the Health
and Retirement Study (HRS) and the Asset and Health Dynamics Among the Oldest-Old (AHEAD)
study. The series currently contains papers on cognitive function, affective function, and physical
function, the latter of which is the focus of this document. The purpose of this document is to aid
prospective analysts in the appropriate use of the physical functioning data. To do so, this document
provides an inventory of the physical functioning measures included in the survey interviews,
describes the formats and origins of these measures, discusses their reliability, validity, and change
over time and several options for working with the evolution of the questions, and provides SAS code
for recoding the variables and creating simple summary measures.
Briefly, the HRS is a national, prospective, probability study designed to enable
characterization of middle-aged people’s health transitions and their effects on resources such as
finances, formal and informal support, family dynamics, and health care. The study began in 1992
(HRS92 or Wave 1) and re-interviewed participants again in 1994) (HRS94 or Wave 2H), 1996
(HRS96 or Wave 3H), 1998 (HRS98 or Wave 4), 2000 (HRS00 or Wave 5), and 2002 (HRS02 or
Wave 6). It surveyed people who were between 51 and 61 years of age in 1992 and their spouses,
irrespective of spouse’s age. The AHEAD study is a companion to the HRS and as such selected
respondents using the same multi-stage area probability sample and screening survey. Its initial
target population was non-institutionalized people 70+ years of age in 1993. Respondents’ spouses
were also interviewed, regardless of whether they were aged 70+ years in 1993. To date, the
interviews have been conducted in 1993/1994 (AHEAD93 or Wave 2A, to be consistent with the
HRS94 or Wave 2H), 1995/1996 (AHEAD95 or Wave 3A, to be consistent with the HRS96 or Wave
3H), 1998 (HRS98 or Wave 4), 2000 (HRS00 or Wave 5), and 2002 (HRS02 or Wave 6). The HRS
and AHEAD merged in 1998 and have remained combined since.
1
For further information on the design and content of the HRS and AHEAD and to obtain the
data, consult the following web site: http://hrsonline.isr.umich.edu. Users must register at this site.
For a second resource on the health measures (as well as many others) in HRS and AHEAD, consult
the RAND Data Documentation for the HRS (St. Clair et al., 2003). It provides data descriptions,
frequencies, and univariate statistics for numerous variables in the HRS data sets 1992 through 2000,
by respondent and respondent’s spouse. The documentation can be used together with the RAND
HRS Data Files, which are available as cross-sectional files or as a longitudinal file with consistent
variable construction and naming conventions across waves.
This report is structured as follows. Section II provides a discussion of why physical
functioning measures were included in the HRS and AHEAD. Section III presents an inventory of
the physical functioning measures for all available waves of HRS and AHEAD. Section IV details
the origins of the measures. Section V outlines several methodological issues with the physical
functioning measures pertinent to the HRS and AHEAD and discusses the comparability of the
measures across waves. Section VI provides the results of investigations into data quality of the
measures. This section includes subsections on internal consistency and measurement properties,
univariate distributions, construct validity, benchmarking against other surveys, and constructed
variables. Section VII discusses certain technicalities of the data, such as changes in the question
wording over time, and options for accommodating these technicalities. It also provides example
SAS code for constructing summary variables.
II. Rationale for measuring physical functioning According to the International Classification of Functioning, Disability and Health (ICF),
physical functioning and disability are a complex, dynamic interaction among the health conditions
of the individual, the environment, and personal factors. Survey questions about physical functioning
2
are a proven critical set of measures that summarize this interaction. Thus, questions about physical
functioning have been included in almost every large survey of older adults, such as the National
Health Interview Survey, Supplement on Aging, the NLTCS, and the Established Populations for
Epidemiologic Studies of the Elderly, etc. (Wiener et al., 1990). They have also been included in
surveys of adults of all ages (e.g., Americans’ Changing Lives), and numerous state-level instruments
intended to predict individuals’ needs for services such as In-Home Aid, Personal Care, and
institutionalization (e.g., the Services and Services Outcome Screen, NC).
The HRS and AHEAD have included a comprehensive battery of physical functioning
measures, in part to summarize the morbidity and environmental interactions within people; however,
there were also other rationales for the inclusion of these measures. One is the evidence that physical
functioning is an important cause and/or outcome of numerous other aspects of people’s well-being,
such as their socioeconomic status (SES) (Bunker, Gomby, & Kehrer; Clark & Maddox, 1992, House
et al., 1994), labor market performance (Ficke, 1992; Martini, 1990; Ries, 1991; Wray, 1996),
affective functioning (Berkman et al., 1986; Blazer et al., 1991; Phifer & Murrell, 1986), and
mortality (Anderson et al. 1998; Wolinsky, Callahan, Fitzgerald, & Johnson, 1993). Of these various
causes and outcomes of physical functioning, the HRS and the AHEAD were designed to provide
researchers with means to study how changes in individuals’ health relates to changes in their labor
force participation, retirement process, economic resources, support services, and family dynamics
(Juster & Suzman, 1995; Soldo, Hurd, Rodgers, & Wallace, 1997). For example, these data can
address such questions as whether people who experience declines in health or catastrophic health
events exit the labor force earlier and/or change the consumption of their economic, health, and social
resources, thereby accelerating the depletion of those resources. Moreover, do people reenter the
labor force and under what circumstances? Does the depletion of such resources contribute to
worsening health or impede recovery?
3
Another rationale for including measures of physical functioning in the HRS and AHEAD
was to address the question of whether functional status in the U.S. population differs over time and
across cohorts. If so, are optimistic claims about the attenuation of disability and disease in later
cohorts warranted (Cf. Fries, 1980)? Some of the work addressing this question is associated with
Manton and colleagues (e.g., Manton, 1990; Manton & Stallard, 1992, 1994; Manton, Corder, &
Stallard, 1993; Manton, Stallard, & Corder, 1998; Manton, Stallard, &Woodbury, 1994;) and their
analyses of the National Long Term Care Survey (NLTCS). Similar to the HRS and AHEAD, the
NLTCS is a nationally representative, prospective survey of older adults. One important difference,
however, is that the main part of the survey, or the “detailed interview”, was designed to characterize
the health problems, cognitive status, formal and informal support, etc. of chronically disabled,
elderly, community-dwelling Americans.1 The HRS and AHEAD do not have this limitation.
Freedman and colleagues (2002) conducted a literature search for other work addressing the
question of changes in functional status in the U.S. population over time. They examined all studies
documenting old age (ages >= 65 or 70 years) disability or functioning trends in the U.S. from the
late 1980s through the 1990s. They identified 16 studies as relevant to their review and examined the
quality, quantity, and consistency of the 8 surveys used in the 16 studies, including AHEAD (through
HRS98). Freedman and colleagues then reported the average annual percent change in functioning
for a sub-set of these studies. HRS and AHEAD can be useful benchmarks for several of the studies
that Freedman and colleagues identified, such as those of local or regional samples (e.g., the
Framingham Heart Study) and/or of specific populations (e.g., Medicare population ages 65 years
and older).
1 The NLTCS contains a “screener” component, which enables researchers to track longitudinal and cohort differences in certain aspects of physical functioning (e.g., activities of daily living), but the “screener” does not contain detailed 4
III. Inventory of measures Physical functioning is a multidimensional concept encompassing mobility, large muscle
functioning, fine motor skills, gross motor skills, and the ability to perform activities of daily living
(ADLs) and instrumental activities of daily living (IADLs). The HRS and AHEAD included as many
questions about these activities as possible so as to facilitate the creation of sensitive, hierarchical
scales that differentiate among people with no problems to severe problems. Also, because there are
disparate points of view on what defines impairment in physical functioning (e.g., needing help to
perform a task vs. self-reported difficulty irrespective of help vs. reliance on equipment, etc.), the
HRS and AHEAD used unfolding questions designed to reveal the nature of impairments as well as
several modules that use different question formats. Here we provide an overview of the measures
included in the interviews and other information pertinent to analysts. We discuss the origins of these
measures in Section IV.
A. Mobility, strength and fine motor skills Table 1 reports the question numbers for the indicators of mobility, strength (large muscle
functioning), gross motor skills, and fine motor skills in the HRS and AHEAD. All of the questions
ask for a self-assessment. While there is some variation in how the questions about these tasks were
asked, the kinds of responses that respondents were prompted for (e.g. yes/no versus none, a little, a
lot, etc.), and follow-up questions, each interview contained questions asking respondents about
whether they have difficulty performing the physical functioning tasks. The question numbers shown
in Table 1 are for these questions about difficulty. Note that due to a change in survey software, the
question numbers in HRS02 and later waves no longer match those of the preceding waves.
information on salient covariates of physical functioning, such as cognition, SES and comorbidities.
5
Table 1
Questions About Mobility, Strength, Gross Motor Skills, and Fine Motor Skills in the HRS and AHEAD
HRS92 HRS94 HRS96 AHEAD93 AHEAD95 HRS98 HRS00 HRS02 Question Core Core Module 3 Core Core Core Core Core Module 10 Core Running/Jogging About 1 Mile B4a B4 B4a E61 E61 E61 E61 HG002Walking Several Blocks B4b B4a B4b E60 E47 E60 E60 E60 M10-13 HG001Walking 1 Block B4c B4b B4c E62 E62 E62 E62 HG003Sitting for About 2 Hours B4e B4d B4e E63 E63 E63 E63 HG004Getting Up From A Chair B4f B4e B4f E64 E64 E64 E64 HG005Climbing Several Flights of Stairs B4h B4g B4h E65 E65 E65 E65 HG006Climbing One Flight of Stairs B4j B4h B4j E66 E48 E66 E66 E66 HG007Lifting or Carrying Over 10 lbs. B4k B4j B4k E70 E50 E70 E70 E70 HG011Stooping, Crouching or Kneeling B4m B4k B4m E67 E67 E67 E67 M10-14 HG008Picking a Dime Up B4n B4m B4n E71 E51 E71 E71 E71 HG012Reaching or Extending Arms B4q B4p B4q E68 E68 E68 E68 HG009Pulling or Pushing Large Objects B4r B4q B4r E69 E49 E69 E69 E69 HG010Notes: This table includes question numbers for questions about respondents' difficulty with each task. Module 10 in HRS00 asks the respondents to report on the functioning of their husbands/wives/partners, rather than themselves.
6
When assessing the prevalence and incidence of difficulties with mobility, strength (large
muscle functioning), gross motor skills, and fine motor skills for the HRS and AHEAD respondents,
it is necessary to be aware of several assumptions the HRS and AHEAD staff made about how these
items relate to each other. Specifically, they assumed that respondents need not be asked about
relatively easy tasks if they reported being able to do more challenging tasks. This assumption is
reflected in the “skip patterns” used in the interviews. For example, respondents who could run or
jog about 1 mile without difficulty were not asked if they could walk several blocks or 1 block, and
respondents who could walk several blocks were not asked if they could walk 1 block (i.e., these
questions were “skipped”). Respondents who could climb several flights of stairs were not asked if
they could climb 1 flight of stairs. Moreover, in AHEAD93, people who had difficulty getting across
a room were not asked if they also had difficulty walking several blocks, climbing a flight of stairs,
and pulling/pushing large objects. In later waves of the HRS and AHEAD, people who had no
problems with mobility, strength, and fine motor skills were not asked about ADLs. The HRS and
AHEAD staff adopted these skip patterns based on their analyses showing that respondents tended to
first lose the ability to perform tasks involving considerable mobility and strength, then the ability to
perform tasks involving moderate mobility and strength, and lastly, the ability to do IADLs and
ADLs. Analysts should consult the codebooks for each wave of the study to determine when skip
patterns were used.
B. Activities of daily living (ADLs) Given its objective to improve understanding of the relationship of health transitions to other
areas of people’s lives, the HRS and AHEAD incorporated a comprehensive set of questions about
activities of daily living (ADLs) (See Tables 2a -2c). First, the surveys included questions about
respondents’ difficulties with ADLs as well as questions about respondents’ help from other people
and equipment in performing these tasks. Second, the surveys contained “core” questions, which
7
were part of the standard interview, and “module” questions, which asked a randomly selected sub-
sample of self-respondents about ADLs using questions from other surveys and/or earlier waves of
the HRS and AHEAD. The modules allow comparison of ADL measures in the core of the HRS and
AHEAD with ADL measures used in other studies (Soldo, Hurd, Rodgers, & Wallace, 1997). They
also enable researchers to assess possible measurement error (or differences) in the core ADL
measures (Freedman, 2000; Rodgers & Miller, 1997).
Table 2a
Questions About ADLs in the HRS and AHEAD
HRS92 HRS94 HRS96 Module B Module C Module D Module 3
Question Core (LSOA) (NLTCS) (Census) Core (HRS92 Core) Core Walking B4d MB1e MC1d B4c B4d E72 Dressing B4t MB1b MC1f B4s B4t E73 Bathing B4p MB1a MC1g B4n B4p E74 Eating B4s MB1c MC1a B4r B4s E75 Getting In/ Out of Bed B4g MB1d MC1b B4f B4g E76 Using the Toilet MB1g MC1h E77 Getting In/ Out of Chairs MC1c Going Outside MB1f MC1e MD3 Taking Care of MD4 Personal Needs Global ADLs Question E79 Notes: Module 1 questions of HRS92 are from the second Longitudinal Study of Aging (LSOA). Module 2 questions of HRS92 are from the National Long Term Care (NLTCS) screen. Module 3 questions of HRS92 are from the Census. Module 1 questions of HRS94 replicated the format used in HRS92; this is because the wording and response categories were modified slightly after HRS92.
8
Table 2b
Questions About ADLs in the HRS and AHEAD
AHEAD93 AHEAD95 HRS98 M3 M4 M5 M1 & 2 M3 M1 & 2 M3 M4 M5 Question Core (LSOA) (NLTCS) (Census) Core (W1 Core) (LSOA) Core (W1 Core) (LSOA) (NLTCS) (Census) Walking E33e M3-6 M4-5 E72 M1-33e M2-33e M3-6 E72 M1-33e M3-6 M4-5Dressing E35b M3-3 M4-7 E73 M1-35b M2-35b M3-3 E73-f M1-35b M3-3 M4-7 Bathing E37b M3-2 M4-8 E74 M1-37b M2-37b M3-2 E74 M1-37b M3-2 M4-8 Eating E39b M3-4 M4-2 E75 M1-39b M2-39b M3-4 E75 M1-39b M3-4 M4-2 Getting In/Out of Bed E43c M3-5 M4-3 E76 M1-43c M2-43c M3-5 E76 M1-43c M3-5 M4-3Using the Toilet E44b M3-7 M4-9 E77 M1-44b M2-44b M3-7 E77 M1-44b M3-7 M4-9 Getting In/Out of Chairs M4-4 M4-4 Going Outside M4-6 M5-6 M4-6 M5-6 Continence M4-10 M4-10 Taking Care of M5-6a M5-6a Personal Needs Global ADLs Question E79 Notes: "M3" refers to "Module 3," "M4" refers to "Module 4," etc. Module 3 questions in all waves are from the second Longitudinal Study of Aging (LSOA). Module 4 questions of AHEAD93 are from the National Long Term Care (NLTCS) screen. Module 5 questions are from the Census. Module 1 and Module 2 questions replicated the format used in AHEAD93; this is because the wording and response categories were modified slightly after AHEAD93.
9
Table 2c
Questions About ADLs in the HRS and AHEAD
HRS00 HRS02 Question Core Core Walking E72 HG016 Dressing E73Y1 HG014 Bathing E74 HG012 Eating E75 HG023 Getting In/Out of Bed E76 HG025 Using the Toilet E77 HG030 Getting In/Out of Chairs Going Outside Continence Taking Care of Personal Needs Global ADLs Question
Similar to the questions about mobility, strength, and fine motor skills, certain questions about
ADLs were “skipped” or omitted during the interview, depending on the respondent’s answers to
preceding questions. These skip patterns changed across interviews as HRS and AHEAD staff
observed the relationships among ADLs and other indicators of physical functioning in this sample.
That is, the ADL questions were asked of all respondents in HRS92-HRS96 and AHEAD93. As of
AHEAD95, however, respondents who reported no difficulty with strength and mobility tasks were
not asked detailed questions about ADLs. Instead, interviewers asked these relatively high-
functioning respondents a global question about whether they have any difficulty with the entire list
of ADLs (“Because of a health problem, do you have any difficulty getting across a room, dressing,
bathing, eating, getting out of bed, or using the toilet?”). Likewise, in HRS98, HRS00, and HRS02,
respondents who either a) had no problems with strength, mobility, and fine motor skills (excluding
running 1 mile or climbing several flights of stairs) or b) reported no more than one problem with
strength, mobility, and fine motor skills (excluding running 1 mile or climbing several flights of
stairs) and no difficulty with dressing were not asked the specific questions about the remaining
10
ADLs. The global ADL question in AHEAD95 was not repeated in later waves because it was found
that almost all respondents who could perform the physical functioning tasks could also perform the
ADLs.
In addition to difficulties with ADL performance, researchers might also be interested in
examining the use of assistance from people and/or equipment, either as an outcome variable (Hartke,
Prohaska, & Furner, 1998), as another way of measuring disability (Agree, 1999), or as a factor in
relieving difficulties (Verbrugge, Rennert, & Madans, 1997). The HRS and AHEAD added questions
about the use of assistance from people for each ADL and about the use of equipment for walking
across a room and getting in and out of bed. (Note: There were several questions about the use of
devices and personal assistance for bathing and money management in Module 7 of AHEAD93 and
AHEAD95 and for bending, lifting, jumping, running, walking around the neighborhood, and fully
using hands and fingers in Module 7 of HRS00.)
C. Instrumental activities of daily living (IADLs) Table 3 documents the questions in the HRS and AHEAD about respondents’ abilities to do
IADLs. Questions about IADLs have been used in many surveys, but there is not a standard list as
there is for ADLs. The HRS and AHEAD asked about grocery shopping, preparing meals, managing
money, making telephone calls, using a calculator, using a microwave, and driving. Limitations in
the latter three activities may manifest before difficulties with other IADLs are evident and/or may
make it difficult to complete other IADLs (e.g. it might be more difficult to grocery shop without
being able to drive). Thus, while these types of activities often may not be included in other surveys,
they were included in the HRS and AHEAD so as to identify the beginning of health and/or cognitive
decline among otherwise high-functioning respondents, particularly the younger HRS respondents.
11
Table 3
Questions About IADLs in the HRS and AHEAD
HRS92 HRS94 HRS96 AHEAD93 AHEAD95 HRS98 HRS00 HRS02 Question Core Core Core Core Core Core Core Module 10 Core
Preparing Hot Meals E95 E52 E95 E95 E95 M10-15 HG041 Shopping for Groceries E96 E52 E96 E96 E96 HG044 Making Telephone Calls B5c E97 E52 E97 E97 E97 HG047 Taking Medications B5d E98 E52 E98 E98 E98 HG050 Managing Money B5b E106 E57 E106 E106 E106 M10-16 HG059 Using a Map B5a B5 E93 E93 E93 E93 HG040 Driving E91 E51c E91 E91 E91 HG037 Using a Microwave B5b Using a Calculator B5c B5a Using a Computer B5d Notes: Module 10 in HRS00 asks the respondents to report on the functioning of their husbands/wives/partners, rather than themselves.
12
Similar to the questions about ADLs, the questions about the IADLs addressed various
aspects of functioning. One aspect was respondents’ self-reported difficulty in performing IADLs.
Another aspect was their dependency, or whether respondents received assistance with the
performance of IADLs. A third aspect was whether respondents who had disabilities in IADLs had
them due to physiological or mental problems (i.e., they don’t or can’t do a task because of a health
problem) or because they did not need to perform those tasks (for example, a spouse might be
responsible for preparing hot meals).
Several skip patterns within the IADL question sequence require note. First, only certain
cohorts of respondents received the questions related to driving; e.g., in HRS96, HRS98, HRS00, and
HRS02, only respondents 68+ (HRS96) and 65+ years of age (HRS98-HRS02) were asked about
driving. Second, respondents were asked if they had any difficulty with IADLs “because of a health
a memory problem.” If they responded that they “can’t do” or “don’t do” an activity, they were again
asked whether their limitation was “because of a health or memory problem.”
D. Accommodations The ways in which older adults experience and interpret their physical functioning
dis/abilities may be related to their use of accommodations (Baltes & Baltes, 1990; Baltes &
Carstensen, 1996). For example, people may use personal assistance and equipment to compensate
for decline. In turn, the use of personal assistance and/or equipment may mitigate further decline or
individuals' perceptions that they are disabled. To address issues such as these, AHEAD93 and
AHEAD95 included Module 7, which gathered information about whether and how respondents used
strategies to assist them with bathing and managing money. Respondents were asked (a) what sorts
of tools they used for bathing (e.g., grab bar, seat or stool, and/or other devices), (b) if someone
actively helped them with bathing (e.g., by helping them get in or out of the bath or shower, and/or
13
wash and dry off), (c) if someone passively helped them with bathing (e.g., by waiting in the house
while they bathed), and (d) if someone helped them manage their bills, savings and investments,
and/or major decisions. The module also contained follow-up questions about the frequency with
which respondents used several of these adjustments and, if someone helped them, who.
Also, HRS00 Module 7 asked several questions about whether respondent had difficulty,
without help or use of equipment of any kind, with a) bending, lifting, jumping, and running, b)
walking around the neighborhood, and c) fully using hands and fingers. For respondents with
difficulty in these domains, a series of follow-up questions asked about whether they usually used
assistive devices, what kind, and whether they relied on other people for assistance.
IV. Pedigree of physical functioning measures
Physical functioning measures like those used in the HRS and AHEAD have a history of use
in clinical settings for diagnostic purposes, in epidemiologic research characterizing patterns of and
risk factors for disability, and in policy research assessing people’s needs for health services. Here
we describe briefly the derivation of the measures.
The use of measures for people’s abilities to lift or carry weight, ascend and descend stairs,
walk, stoop, bend, or kneel, reach, and use hands and fingers gained widespread use after appearing
in work by Rosow and Breslau (1966) and Nagi (1969, 1976). Rosow and Breslau measured older
people’s abilities to walk about one-half mile, plus other physical activities like heavy housework,
and created a Guttman Scale of physical health that they then used in their examination of older
adults’ social participation. Nagi used measures of physical functioning, combined with measures of
ADLs and certain IADLs, to develop a scheme differentiating among people with no difficulties,
people with some difficulties but independent, people needing assistance in mobility outside of the
home (these people also need help with housekeeping, work, and shopping) and people needing
14
assistance in personal care (Nagi, 1976). In a more recent framework (“disablement process”)
proposed by Verbrugge and Jette (1994), losses of mobility, strength, and fine motor skills were
defined as “impairment.” Verbrugge and Jette hypothesized that impairment usually (although not
inevitably) follows pathology and precedes “disability.” They defined disability as individual’s
inability to work and fulfill social roles.
Katz and associates (Katz et al., 1963) developed the first scale of ADLs now used in the HRS
and AHEAD surveys, in order to provide a way of assessing the effectiveness of treatments for older
people with hip fractures. Given its usefulness for differentiating among people in this specific
population, Katz and associates perceived that it would be a helpful instrument for understanding the
course of decline in basic functioning [i.e., “activities which people perform habitually and
universally” (p. 94)] among well people also. Their scale covered bathing, dressing, going to the
toilet, transfer in and out of chairs, continence, and feeding. It involved rating individuals’ “adequacy
of performance” (p. 94) in these tasks and their stages of functional loss (e.g., generally people first
lose the ability to bathe independently, then dressing, etc.).2 Since appearing in Katz et al.’s work,
ADLs have been used extensively and in various combinations in health-related research.
The IADLs in the HRS and AHEAD borrowed from the inventory developed by Lawton and
Brody (1969). In Lawton and Brody’s 1969 framework, there are multiple, general areas of human
functioning that differ in terms of complexity. In order from least to most complex, these areas are
life maintenance, perception-cognition, physical self-care, instrumental self-care, effectance, and
social behavior. Lawton and Brody created the IADL scale to assess the instrumental self-care area
of functioning, which they asserted had been measured inadequately up to that point. This scale
included the ability to use the phone, shop, prepare food, clean house, launder clothing, use
2 Katz et al.’s (1963) scheme is as follows: a) independent in all activities; b) independent in all except one activity; c) independent in all activities except bathing and one additional activity; d) independent in all activities except bathing, dressing, and one additional activity; e) independent in all activities except bathing, dressing, going to toilet, and one
15
transportation, take medications, and handle finances. The range of items was explicitly designed to
tap functioning of both men and women, who may perform different instrumental tasks, and older
people living in a variety of settings, such as homes for aged people, foster homes, the community,
etc. Lawton and Brody validated their scale among people living in numerous situations and it has
been in wide use since.
V. Special methodological issues Two methodological issues pertinent to the HRS and AHEAD surveys are discussed here—
measurement error and change in question wording/sequencing over time.
A. Measurement error An underlying assumption of much research using measures of physical functioning is that the
observed variables (e.g., self-reported ability to walk, eat, climb stairs, etc.) correlate perfectly with
their corresponding latent variables or concepts (Bollen, 1989). This assumption may be erroneous,
however; e.g., Jette’s (1994) study of ADL disability found that its prevalence differed widely
according to how the questions about ADLs were asked. Prevalence was about five times greater for
questions about difficulty compared to questions about human assistance. From this finding we can
infer that, although each ADL question was intended to be a perfect indicator of a latent variable
representing disability in a specific task, some were more or less perfect indicators than others.
Which types of indicators are most accurate remains an open question.
Inasmuch as observed variables do not correlate perfectly with corresponding latent variables
in the HRS and AHEAD, measurement error exists. This is a problem that the HRS and AHEAD
share with most other studies. Yet, with their experimental modules containing differently phrased
ADL questions, the HRS and AHEAD provide a unique opportunity to examine the extent of
additional activity; f) independent in all except bathing, dressing, going to toilet, transferring, and one additional function; g) dependent in all activities; and other (dependent in at least 2 activities, but classifiable as C, D, E, or F). 16
measurement error. Rodgers and Miller (1997) took such an opportunity by cross-classifying
respondents’ answers to ADL questions in AHEAD93 of the core with their answers to ADL
questions in two modules of AHEAD93. One module contained ADL questions based on the format
used in the National Health Interview Survey’s Supplement on Aging (LSOA) and the other
contained questions based on the format of the National Long Term Care Survey’s (NLTCS)
screener. Rodgers and Miller found large discrepancies in the proportion of people with ADL
difficulties as reported using these three question formats. Kappa agreement coefficients for each
ADL showed that consistency between answers to the core and module questions was low for the
core/LSOA module comparison and moderate for the core/NLTCS module comparison. This finding
suggested measurement error. Finally, by correlating other health-related variables with various help
and difficulty limitation scales (observed separately) based on (1) respondents' answers to the core
questions only, (2) respondents' answers to the module questions only, (3) respondents' answers to the
core or module questions, and (4) respondents' answers to the core and module questions, Rodgers
and Miller (1997) ascertained that the measurement error is in the direction of underreporting of
ADL difficulties. In other words, respondents with ADL-related problems may have failed to report
them, but individuals without problems rarely reported having them. An additional and as yet
uninvestigated methodological issue is the extent to which respondents' functional transitions are
“true change” or are due to measurement error in the ADL items over time.
B. Comparison of questions across waves Analyses of transitions and trajectories using longitudinal data have become increasingly
common in social science research. Transitions refer to individual and group short-term, discrete
changes from one state to another (e.g., from unimpaired to impaired, from working to retirement,
etc.) (George, 1993). Trajectories refer to long-term patterns of stability and change (sequences of
transitions) (Elder, 1985; George, 1993). An important prerequisite to observing transitions and
17
trajectories is repeated measures for the same individuals. To avoid confounding measurement error
with substantive change, another prerequisite is to use the same questions at each observation of a
longitudinal survey (Cook & Campbell, 1979).
With respect to physical functioning, the HRS and AHEAD facilitate comparison over time
by covering the same tasks in every interview, rarely eliminating questions and often adding
questions. In addition, the HRS and AHEAD followed all respondents over time irrespective of
whether they continued to live in the community. The HRS and AHEAD also interviewed relatives
of respondents who died between waves in what is called the “Exit Interview”. The Exit Interview
contains information about deceased respondents’ health (including physical functioning) and health
care expenditures prior to death and the disposition of assets.
But there are several ways in which the HRS and AHEAD could make comparison over time
less straightforward than might be desirable. First, the sequence of ADL-related questions changed
between AHEAD93 and AHEAD95. That is, respondents were asked if they had any help
performing each task and, if so, the frequency of help and who helped. For the tasks of walking
across a room and getting in and out of bed, the interviewers also asked respondents if they used
equipment or devices and, if so, what kind of equipment and how often they used it. Then, for each
ADL, respondents were asked if they had any difficulty (See Figure 1). In AHEAD95, the difficulty
question was first in the sequence, as opposed to second or third (See Figure 2). Fortunately, later
waves maintained consistency in the sequencing of the ADL-related questions. Second, the wording
of the functional status questions changed over time. To illustrate, AHEAD93 ADL-related questions
asked, “Does anyone ever help you...?” and “Do you have difficulty...?”. For the AHEAD93 IADLs,
the questions began with, "Are you able to...?" But questions about ADL functioning in AHEAD95
began with, "Because of a health or memory problem, do you have any difficulty...?" Table 4 shows
18
the lead-in wording for questions about physical functioning in the core and modules of HRS and
AHEAD through HRS02.
Good rationales existed for the change in question sequence and wording. The questions in
AHEAD93 were designed to address elders’ levels of help, consistent with the overarching goals of
the HRS and AHEAD study. These questions differed from those used in other surveys of older
adults, however, which researchers argued made cross-study comparisons difficult. Moreover, the
help-oriented format was not as effective at addressing the issue of elders’ actual difficulty with
physical functioning tasks. Finally, the change in question format between AHEAD93 and
AHEAD95 was in preparation for the merging of HRS and AHEAD as of HRS98. Strategies for
accommodating changes in question wording and format are discussed later in this document.
19
Figure 1
Question flow of ADLs in the core of AHEAD93
Does anyone ever help you get across a room?
No Yes Don’t Do (Skip to next ADL) Do you get that help
most of the time, some of the time, or only occasionally?
Who helps?
Do you usually use that equipment?
No Yes
What equipment is that?
Do you ever use equipment or devices (e.g., cane) when crossing a room?
Yes
Is that a little or a lot of difficulty?
(Even when someone helps/ use equipment/ without help) Do you have any difficulty walking across the room?
No
Next ADL. Next ADL.
20
Figure 2
Question flow of ADLs in the core of AHEAD95
Because of a health or memory problem, do you have any difficulty walking across the room?
What equipment is that?
Do you ever use equipment or devices such (e.g., cane) when crossing a room?
No Y
YesNo Next ADL.
(If “yes” to the question about difficulty) Does anyone ever help you get across a room?
Next ADL.
21
Table 4
Sample Lead-in Questions about Physical Functioning in the HRS Study
Core Questions by Wave Mobility, Strength, & Fine Motor SkillsHRS92: How difficult is it for you to…? HRS94: Do you have any difficulty….? HRS96: Because of a health problem do you have any difficulty with…? AHEAD93: Do you have any difficulty….? AHEAD95: Because of a health problem do you have any difficulty with…?
HRS98: Because of a health problem do you have any difficulty with…? HRS00 Because of a health problem do you have any difficulty with…? HRS02 Because of a health problem do you have any difficulty with…? ADLs HRS92: How difficult is it for you to…? HRS94: Do you have any difficulty….? HRS96: Because of a health problem do you have any difficulty with…? AHEAD93: Does anyone ever help you to…? (followed by questions about difficulty) AHEAD95: Because of a health problem or memory problem do you have any difficulty with…? HRS98: Because of a health problem or memory problem do you have any difficulty with…?
HRS00 Because of a health problem or memory problem do you have any difficulty with…? HRS02 Because of a health problem or memory problem do you have any difficulty with…? IADLs HRS92: How difficult is it for you to…? HRS94: Do you have any difficulty….? HRS96: Because of a health problem do you have any difficulty with…? AHEAD93: Are you able to (activity) without help? AHEAD95: Because of a health problem or memory problem do you have any difficulty with…? HRS98: Because of a health problem or memory problem do you have any difficulty with…? HRS00 Because of a health problem or memory problem do you have any difficulty with…? HRS02 Because of a health problem or memory problem do you have any difficulty with…?
Module Questions by Wave Mobility, Strength, & Fine Motor SkillsHRS94: How difficult is it for you to…? (Module 3)
ADLs HRS92: Because of a health or physical problem, do you have any difficulty…? (Module B)
Do you have any problem (activity) without the help of another person or special equipment? (Module C)
Do you have any difficulty taking care of your own personal needs, such as bathing, etc.? (Module D)
HRS94: How difficult is it for you to…? (Module 3) AHEAD93: Because of a health or physical problem, do you have any difficulty…? (Module 3)
Do you have any problem (activity) without the help of another person or special equipment? (Module 4)
22
Do you have any difficulty taking care of your own personal needs, such as bathing, etc.? (Module 5)
AHEAD95: Does anyone ever help you to…? (followed by questions about difficulty) (Modules 1 & 2) Because of a health or physical problem, do you have any difficulty…? (Module 3) HRS98: Does anyone ever help you to…? (followed by questions about difficulty) (Modules 1 & 2) Because of a health or physical problem, do you have any difficulty…? (Module 3)
Do you have any problem (activity) without the help of another person or special equipment? (Module 4)
Do you have any difficulty taking care of your own personal needs, such as bathing, etc.? (Module 5)
IADLs
HRS00 Because of a health or memory problem, does he/she (husband/wife/partner) have any difficulty…? (Module 10)
VI. Evaluation of the measures and data quality
A. Previous research on internal consistency and measurement properties Recent debates on the measurement of physical functioning have raised questions about
whether strength, mobility, fine motor skills, ADL ability, and IADL ability comprise one
comprehensive domain or multiple, related, hierarchical domains. According to studies by Wolinsky
and Johnson (1992), Johnson and Wolinsky (1993), and others (e.g., Clark, Stump, & Wolinksy,
1997; Fitzgerald et al., 1993), strength and mobility, ADLs, and IADLs represent five areas of
functioning for respondents of the LSOA: lower and upper body functioning and basic, household,
and advanced ADLs. Lower body functioning is comprised of such tasks as stooping, kneeling or
crouching and walking several blocks. Upper body functioning consists of tasks like reaching over
one's head. Basic ADLs are bathing, dressing, transfer in and out of chairs and bed, walking, and
toileting. Household ADLs include shopping, meal preparation, and housework, and advanced ADLs
are those that may reflect cognitive functioning, such as managing money, using the telephone, taking
medications and eating. Wolinsky and colleagues argue that it is important to examine these
functional domains separately, either as causes or outcomes.
Spector and Fleishman (1998) provide a counter perspective to that of Wolinsky and
colleagues. They argue that combining IADL and ADL items in one scale could increase the range
23
and sensitivity of these items to measure functional impairment. Using data from the 1989 National
Long-Term Care Survey, they performed factor analyses using tetrachoric correlations of IADL and
ADL measures. The measures in their analyses included going outside of walking distance,
shopping, doing laundry, preparing meals, taking medicines, finances, light housework, telephoning,
getting around outside, bathing, getting around inside, dressing, transferring, toileting, help with
incontinence, and feeding. Their study found that all but one of the IADL and ADL measures
reflected one dimension. Next, Spector and Fleishman used item response theory (IRT) methods to
determine the best psychometric approach for combining the IADL and ADL measures into scales;
i.e., they examined the fit of a one-parameter model versus a two-parameter model. From this
analysis, Spector and Fleishman conclude that the one-parameter model was most appropriate,
meaning that overall levels of functional impairment could be measured simply by identifying the
total number of items with which respondents reported difficulty. Had a two-parameter model been
more appropriate, the “correct” way to measure functional impairment would have involved
consideration of the number, combination, and ordering of activities with which respondents have
difficulty.
To our knowledge, two studies to date have looked at internal consistency and/or
measurement properties of the physical functioning measures in the context of the HRS and AHEAD.
Wallace and Herzog (1995) performed an exploratory factor analysis of the physical functioning
items in HRS92 and identified three domains: 1) mobility, including all ambulation items except
jogging and climbing stairs (i.e., mobility difficulty index); 2) lower and upper body strength (i.e.,
large muscle difficulty index); and 3) ADLs (ADL difficulty index). Using AHEAD93, Stump and
associates (1997) attempted to confirm the multi-dimensional structure identified by Wolinksy and
Johnson (1992). They found that the physical functioning measures formed factors around lower
body disability and basic, household, and advanced ADLs. The lower body items included walking
24
several blocks, climbing one flight of stairs, pulling and pushing large objects, and carrying 10
pounds. The basic ADL items were dressing, bathing, getting in and out of bed, and using the toilet.
The household ADLs were preparing hot meals and shopping for groceries and the advanced ADLs
were managing money, making telephone calls, and taking medication.
B. A basic examination of internal consistency and measurement properties This section reports on a systematic examination of the reliability and factor structure of the
physical functioning measures included in all waves of the HRS and AHEAD through HRS00. [This
information is not shown for HRS02, as these data are in “early release” form (version 1.0) as of the
publication of this document and may change slightly during the data cleaning process.] For each
wave, we calculated the correlation matrix and Cronbach’s coefficient alpha for all of the physical
functioning measures together. Next, we conducted exploratory factor analyses of the physical
functioning measures to examine whether they represented multiple domains of functioning, such as
lower body functioning, basic ADL functioning, etc. Third, we created additive scales for each
domain suggested by the factor analysis and assessed the Cronbach’s coefficient alphas for these.
Arguably, there are alternative methods for examining internal consistency and measurement
properties than we used for this documentation. This documentation, however, is meant to illustrate
the quality of the physical functioning variables in the HRS and AHEAD, not to prescribe how they
should be used in research.
Table 5 shows the Cronbach’s coefficient alphas for all of the physical functioning measures
together (Column 2) and by conceptual domain (Columns 3-5). It appears from this table that the
different physical functioning scales were fairly reliable; i.e., the Cronbach’s coefficient alphas
generally exceeded the minimum value of .70 suggested by Nunnally (1978) and were as high as .92
in some instances. One exception is the IADL summated scale in HRS92, which contained
experimental IADL items including using a map, calculator, computer, and microwave. These
25
experimental items were added by Drs. Robert Wallace and A. Regula Herzog. The Cronbach’s
coefficient alpha of the scale with these items was .60.
Tables 6a and 6b show the results from exploratory factor analyses of the physical functioning
variables in the HRS and AHEAD. We used an oblique rotation method (Promax), which assumes
that the underlying, latent domains are correlated. Indeed, the inter-factor correlations were
moderately high in most cases. The correlation between the factor for strength/mobility and the
factor for ADLs ranges from .44 (for AHEAD93 and AHEAD95) to .48 (HRS92 -HRS96 and
HRS98). The correlation between the factor for ADLs and the factor for IADLs ranged from .19 (for
HRS96) to .55 (HRS98). The correlation between the factor for strength/mobility and the factors for
IADLs ranged from .26 to .44. These results suggest that the variables might be indicators for a
latent, second-order factor of overall physical functioning, yet they are distinct enough from each
other to justify grouping them as we did in some of the analyses of internal consistency; i.e., one
group was comprised of strength/mobility measures, another of ADLs, and a third of IADLs.
Table 5 Analysis of Internal Consistency Reliability for the Physical Functioning Measures in the HRS and AHEAD
(Cronbach's Alphas) All Physical Strength, Mobility, & Functioning Measuresa Motor Skills ADLs IADLsa
HRS92 .87/.86 .85 .77 .60b
HRS94 .89/.89 .87 .78 .69/.69 HRS96 .88/.88 .86 .78 .71/.61 AHEAD93 .90/.90 .81 .82 .78/.78 AHEAD95 .92/.92 .85 .85 .85/.82 HRS98 .91/.92 .87 .84 .83/.83 HRS00 .92/.92 .87 .84 .85/.86 Notes: The analytic sample consists of age-eligible respondents with complete information on physical functioning. All physical functioning variables were dichotomized (no difficulty vs. any difficulty) for consistency. See Tables 1-3 for a listing of the specific measures available for each category per wave. a The first alpha in each cell is from analyses omitting atypical measures (e.g., driving). The second includes them.b HRS92 only included measures of people's difficulties using maps, microwaves, calculators, and computers.
26
The exploratory factor analyses also indicated that three latent factors accounted for
covariation in the data on physical functioning in both the HRS and AHEAD and that the physical
functioning variables corresponded to (or “loaded on”) the three factors assumed a priori when
developing the HRS/AHEAD and organizing this document -- strength, mobility and motor skills,
ADLs, and IADLs.3 Using 40 as the critical minimum value for the standardized regression
coefficients (Hatcher, 1998), the measures of jogging, walking several blocks, getting up from a
chair, climbing stairs, lifting or carrying, and pulling or pushing all loaded on the factor representing
mobility and strength in HRS92, AHEAD93, etc. Walking across a room, dressing, eating, bathing,
and getting in/out of bed loaded on the factor representing ADLs in HRS92, HRS98, and so forth.
Additionally, the physical functioning variables usually had high loadings for only one factor.
But the exploratory factor analyses also pointed to several differences in how the HRS and
AHEAD physical functioning measures relate to the assumed latent domains. For instance, the
measure for “picking a dime up” did not load on any factor in HRS94-HRS96 and loaded on the ADL
factor in HRS92, AHEAD93-AHEAD95, HRS98 and HRS00 rather than the factor representing
strength, mobility, and motor skills. Climbing one flight of stairs in HRS94, shopping for groceries
in AHEAD93, stooping and getting in/out of bed in AHEAD95, and driving in AHEAD93 and
AHEAD95 loaded on multiple factors, rather than just one. Moreover, the results for HRS00 differed
from those of the other waves, with several of the standard ADL measures having high standardized
regression coefficients for both the ADL and IADL factors. In sum, analysts need to be aware of the
similarities and differences in how the HRS and AHEAD physical functioning measures relate to the
assumed latent domains and examine them further before constructing scales.
3 These analyses specified that three factors be retained and rotated. We did this so as to be consistent with the three factors commonly referred to in the gerontological and health literature: 1) strength, mobility, and motor skills; 2) ADLs; and 3) IADLs. Analysts may wish to allow their statistical program to freely determine the number of factors. 27
We did not conduct similar analyses of the physical functioning measures in the experimental
modules because the modules only contained questions about tasks likely to fall in the ADL domain.
Confirmation of this expectation might be an objective for future research.
C. Univariate distributions This section presents the univariate distributions of physical functioning scales in the HRS
and AHEAD. We constructed simple scales for mobility, strength, and fine motor skills4 (Table 7),
ADLs (Table 8), and IADLs (Table 9). We did this by summing respondents’ answers about whether
they had difficulty performing these tasks; lower scores indicated better functioning and higher scores
indicated worse functioning. Review Tables 1 – 3 in this documentation for guidance on which
activities comprised each scale (excluding the global question about ADLs). For the most part, these
scales were consistent with the results for the examination of internal consistency and measurement
properties of the physical functioning measures.
4 One “fine motor skill”—“picking up a dime”—did not load on the overall domain of mobility, strength, and fine motor skills. Nonetheless, we added this item to the summated scale shown in Tables 7 because many users might have theoretical reasons for including this task with their own scales tapping mobility, strength, and motor skills. 28
Table 6a
Exploratory Factor Analysis of the Physical Functioning Measures in the HRS
HRS92 (n = 9,824) HRS94 (n = 8,564) HRS96 (n = 8,129 ) Factor 1 Factor 2 Factor 3 Factor 1 Factor 2 Factor 3 Factor 1 Factor 2 Factor 3Jogging About 1 Mile 53 -22 1 60 -14 -8 55 -25 10 Walking Several Blocks 66 16 -5 63 30 -10 67 19 -4 Walking 1 Block 39 49 -7 32 63 -10 39 41 -8 Sitting for About 2 Hours 35 10 8 51 -6 24 60 2 -2 Getting Up From A Chair 65 1 0 68 -8 7 70 -1 0 Climbing Several Flights 75 -14 3 77 -4 -4 75 -13 8 Climbing 1 Flight 60 26 0 50 42 -12 52 28 -4 Lifting or Carrying 64 12 2 66 10 7 60 17 5 Stooping, Crouching, etc. 75 -9 2 74 -9 5 72 -2 -2 Picking a Dime Up 6 47 9 7 38 30 13 26 19 Reaching or Extending 31 39 7 42 19 20 43 22 3 Pulling or Pushing 66 10 4 66 8 9 64 10 3 Walking 9 72 -9 8 71 7 3 69 -4 Dressing -14 85 0 -8 80 2 14 58 -1 Bathing 1 80 0 1 77 -5 -7 72 5 Eating -25 68 7 -24 67 7 -14 50 16 Getting In/ Out of Bed 24 54 -2 12 56 19 13 60 -5 Using the Toilet -1 71 -18 Preparing Hot Meals -12 56 33 Shopping for Groceries 7 54 25 Making Telephone Calls -9 31 61 1 3 69Taking Medications -11 27 58 4 1 62Managing Money -6 11 66 -2 15 64Using a Map 14 -7 64 20 -22 65 29 -18 39 Using a Microwave -7 11 66 Using a Calculator -3 9 74 11 -6 69 Using a Computer 6 8 60 Correlation between Factors 1 & 2 0.48 0.48 0.48 Correlation between Factors 1 & 3 0.26 0.36 0.34 Correlation between Factors 2 & 3 0.24 0.46 0.19 Eigenvalue 6.4 1.7 1.6 7.7 2 1.3 7 2 1.3 Percent of variance explained 30.7% 8.3% 7.5% 34.9% 9.2% 5.9% 29.1% 8.4% 5.2% Notes: This table reports the standardized regression coefficients from the Promax (oblique) rotated factor patterns. All factor loadings were rounded and multiplied by 100. The sample consisted of age-eligible respondents with complete information on physical functioning. In HRS96, only certain older cohorts were asked about driving ability. This means that a substantial number of cases were excluded from the analyses. However, dropping the driving variable does not alter the general factor pattern. Driving was not asked about in earlier waves. All physical functioning variables were dichotomized (no difficulty vs. any difficulty).
29
Table 6b
Exploratory Factor Analysis of the Physical Functioning Measures in HRS and AHEAD
AHEAD93 (n = 7,410 )
AHEAD95 (n = 6,263 ) HRS98 (n = 11,014 ) HRS00 (n = 18,532 )
Factor 1 Factor 2 Factor 3 Factor 1 Factor 2 Factor 3 Factor 1 Factor 2 Factor 3 Factor 1 Factor 2 Factor 3Jogging About 1 Mile -11 70 12 4 52 -24 -11 74 7 Walking Several Blocks 4 70 12 5 70 10 5 74 7 Walking 1 Block 77 -4 6 27 49 15 14 46 31 26 52 13 Sitting for About 2 Hours -24 7 65 -29 27 42 -24 11 62Getting Up From A Chair -14 35 55 -23 47 43 -19 37 54Climbing Several Flights 75 -3 10 -7 74 1 5 77 -12 -6 77 -1 Climbing 1 Flight 17 61 8 15 60 14 19 64 5 Lifting or Carrying 79 -2 1 13 61 9 17 61 7 15 62 7 Stooping, Crouching, etc. -16 48 43 -16 59 25 -15 54 34 Picking a Dime Up 8 -13 55 18 -14 54 -3 -1 57 18 -9 53Reaching or Extending -1 21 45 -9 26 43 2 16 48Pulling or Pushing 77 -5 6 7 60 12 13 62 5 9 58 12 Walking 60 1 26 44 16 29 19 10 57 46 14 29Dressing 22 6 63 38 3 47 15 3 64 38 0 48Bathing 26 22 46 58 10 22 36 7 48 58 8 25Eating -10 36 53 65 -18 28 37 -18 53 62 -17 27Getting In/ Out of Bed 21 -1 65 41 -6 48 13 -7 72 41 -10 53Using the Toilet -5 4 77 38 -9 50 11 -5 70 38 -9 51Preparing Hot Meals 8 58 25 78 7 0 69 -1 20 81 5 -1Shopping for Groceries 44 52 -1 67 28 -7 60 17 17 71 21 -3Making Telephone Calls -19 73 17 79 -12 1 70 -9 9 78 -8 -4Taking Medications -18 74 20 75 -14 4 71 -11 8 76 -9 -4Managing Money 11 74 -17 80 4 -12 81 0 -2 84 3 -13Using a Map 43 7 -13 56 11 -17Using a Microwave Using a Calculator Using a Computer
30
Driving 44 52 -1 52 44 31 60 31 -15 56 38 -26 Correlation between Factors 1 & 2 0.40 0.44 0.37 0.44 Correlation between Factors 1 & 3 0.44 0.47 0.55 0.45 Correlation between Factors 2 & 3 0.50 0.44 0.48 0.47 Eigenvalue 6.9 1.7 1.1 9.0 2.1 1.3 9.0 2.1 1.2 9.0 2.2 1.2Percent of variance explained 40.6% 10.0% 6.4% 36.0% 8.2% 5.1% 36.2% 8.6% 4.8% 37.4% 9.0% 5.0%Notes: This table reports the standardized regression coefficients from the Promax (oblique) rotated factor patterns. All factor loadings were rounded and multiplied by 100. The sample consisted of age-eligible respondents with complete information on physical functioning. In HRS98 and HRS00, only certain older cohorts were asked about driving ability. This means that a substantial number of cases were excluded from the analyses. However, dropping the driving variable does not alter the general factor pattern. All physical functioning variables were dichotomized (no difficulty vs. any difficulty).
31
There are several notable trends regarding the distribution of physical functioning difficulties
in the HRS and AHEAD. First, respondents had more difficulties with mobility, strength, and fine
motor skills than with ADLs or IADLs. ADL impairments were relatively uncommon and the
univariate distributions were highly skewed. IADL impairments were more common than ADL
impairments, but less so than difficulties with mobility, strength, and fine motor skills. Second,
despite probable differences due to changes in the questions, it was clear that physical functioning
problems were more common for the older AHEAD respondents. This can be ascertained from a
comparison of HRS94 and AHEAD95, which used fairly congruent questions; e.g., 19.1% of HRS94
respondents had no difficulties with mobility, strength, and fine motor skills, compared to 3.3% of
AHEAD95 respondents. Unfortunately, additional detailed conclusions about the distributions of
impairment over time and between the HRS and AHEAD were difficult due to cross-wave
differences in question wording and cross-study differences in the number of items included in
certain waves.
32
Table 7
Distribution of Limitations in Mobility, Strength, and Motor Skills in the HRS and AHEAD
HRS92 HRS94 HRS96 AHEAD93 AHEAD95 HRS98 HRS00SCORE (n = 9,824 ) (n = 8,970 ) (n = 8,366) (n = 7,447 ) (n= 6,307 ) (n = 20,432 ) (n = 18,532 )
Range 0-12 0-12 0-12 0-5 0-12 0-12 0-12Total (%) 100.0 100.0 100.0 100.0 100.0 100.0 100.0
0 8.2 19.1 15.0 45.2 3.3 9.8 9.11 21.1 23.8 26.0 15.4 14.6 21.5 20.92 17.0 15.4 15.8 11.4 12.5 14.7 14.33 12.6 10.0 10.5 9.6 11.3 11.2 11.04 9.6 7.1 7.3 14.1 9.5 8.4 9.05 7.4 5.3 5.4 4.2 8.3 6.9 7.16 5.2 4.4 4.5 NA 7.9 5.5 5.77 4.6 3.3 4.0 NA 7.3 4.9 4.98 3.6 2.8 3.4 NA 6.6 4.7 5.09 3.5 2.8 3.1 NA 7.1 4.4 4.9
10 3.2 2.5 2.7 NA 6.2 3.9 4.111 2.6 1.7 1.6 NA 3.8 2.8 3.012 1.5 1.7 .6 NA 1.6 1.2 1.1
Notes: Samples included age-eligible respondents only with complete information on mobility, strength, and motor skills. NA means not applicable.
33
Table 8
Distribution of ADL Limitations in the HRS and AHEAD
HRS92 HRS94 HRS96 AHEAD93 AHEAD95 HRS98 HRS00SCORE (n = 9,824 ) (n = 8,970 ) (n = 8,366) (n = 7,447 ) (n = 6,307 ) (n = 20,432 ) (n =18,531 )
Range 0-5 0-5 0-6 0-6 0-6 0-6 0-6Total (%) 100.0 100.0 100.0 100.0 100.0 100.0 100.0
0 88.9 90.9 87.9 69.5 70.3 81.8 80.91 6.0 4.4 5.9 13.4 11.9 7.7 8.52 2.2 2.3 2.6 6.4 6.6 3.8 3.83 1.4 1.1 1.5 4.0 3.5 2.4 2.44 1.0 .8 1.1 2.6 2.3 1.4 1.55 .5 .5 .5 1.9 1.9 1.3 1.36 NA NA .4 2.2 3.6 1.6 1.6
Notes: Samples included age-eligible respondents only with complete information on ADLs. NA means not applicable.
34
Table 9
Distribution of IADL Limitations in the HRS and AHEAD
HRS92 HRS94 HRS96 AHEAD93 AHEAD95 HRS98 HRS00SCORE (n = 9,824 ) (n = 8,970 ) (n = 8,366) (n = 7,447 ) (n= 6,307 ) (n = 20,432 ) (n = 18,506 )
Range 0-4 0-5 0-5 0-5 0-5 0-5 0-5Total (%) 100.0 100.0 100.0 100.0 100.0 100.0 100.0
0 24.0 68.7 90.6 71.2 73.5 84.7 84.61 36.6 20.2 5.5 14.8 10.4 7.2 6.82 21.7 6.3 2.2 6.2 5.4 3.1 3.23 11.0 2.1 .9 3.2 3.4 1.8 1.84 6.6 .9 .5 2.0 3.3 1.5 1.65 X 1.9 .4 2.7 3.9 1.8 2.0
Notes: Samples included age-eligible respondents only with complete information on IADLs. NA means not applicable. The univariate distributions for HRS92 and HRS94 contained IADL items that are not considered "standard"; i.e., using a computer, microwave, etc.
35
D. Previous research on construct validity This section reviews previous research on the construct validity of the physical functioning
measures included in the HRS and AHEAD. Construct validity refers to how well a scale reflects the
underlying concept it is intended or believed to measure. It is believed to reflect that underlying
concept well if the observed scale relates to other characteristics of the respondent in ways that theory
would predict them to. For example, if the physical functioning measures were valid, they should
have had strong bivariate and multivariate relationships with age, health conditions, self-rated health,
etc.
To our knowledge, there are only three published studies that have explicitly addressed
construct validity in the physical functioning measures of the HRS and AHEAD. The first is a study
by Wallace and Herzog (1995) that assessed the construct validity of these domains by juxtaposing
respondents' level of difficulty in each against measures of disease included in HRS92. As expected,
they found that individuals with health conditions such as diabetes, cancer, heart problems and stroke,
had more problems with mobility and ADLs. The second study, by Rodgers and Miller (1997),
examined the extent to which three different summary scores representing AHEAD93 respondents'
ADL functioning were correlated with several criterion variables such as age, health rating, number
of health conditions, and number of doctor visits. Rodgers and Miller obtained the first score by
regressing the criteria variables on respondents’ answers to ADL questions about getting help,
amount of help received, having difficulty, and amount of difficulty and then taking the average
values of the regression coefficients to create a composite score. The other two scores were additive
scales of respondents' limitations in all six ADLs (measured in terms of whether respondents have
help, difficulty, or use equipment), and respondents' limitations in three, principal activities (i.e.,
getting across the room, dressing, and bathing). All correlations were in the expected direction and
statistically significant, providing evidence of moderate to high construct validity. With regard to
36
differences among the different summary measures that Rodgers and Miller (1997) created, the two
additive scales show slightly higher construct validity than the composite score. The third study, by
Stump and colleagues (1997), related basic ADL functioning (see section V.A.), household ADL
functioning, advanced ADL functioning, and lower body functioning among AHEAD93 respondents
to other aspects of health and social-demographics, including perceived health, affective functioning,
health conditions, gender, race, and age. As the researchers expected, most of the health conditions
were associated with worse functioning, as outlined by Stump and associates. Several of the physical
functioning domains were related to poorer affective functioning and perceived health, such as basic
ADL functioning and lower body functioning.
In addition to the three studies described above, many studies focusing on issues other than
construct validity per se but controlling for physical functioning have established the validity of these
measures almost unequivocally. Such studies include an analysis of gender and disability using
HRS92 and AHEAD93 (Wray & Blaum, 2001), worsening depressive symptoms and its competing
risks (i.e., death, lost to follow-up) in AHEAD93-HRS98 (Fonda, Herzog, & Wallace, 2001),
race/ethnicity differences in the transition from health to disability in AHEAD93 (Zsembik, Peek, &
Peek, 2000), etc. Refer to the following web sites for a list of additional studies:
http://www.umich.edu/~hrswww/pubs/biblio.html. See also reports presented at the symposium
titled “Co-morbidity in older age: Findings from the Health and Retirement Study” at the 2001
meeting of the Gerontological Society of America. Since there are so many studies available that
confirm the construct validity of the physical functioning measures in the HRS and AHEAD, we will
not provide an analysis of construct validity in this documentation.
F. Benchmarking against other surveys and prevalence estimates One method for evaluating the quality of the physical functioning measures is to compare
prevalence rates of difficulty or disability against rates found in similar analyses of other data. We
37
http://www.umich.edu/~hrswww/pubs/biblio.html
call this type of comparison “benchmarking.” Benchmarking was not possible for the HRS because
there are no other nationally representative surveys assessing the health of people between the ages of
51 and 61 years. However, two data sources provided reasonable benchmarks for the AHEAD--the
Supplement on Aging (SOA) II (1995) and the National Medical Expenditures Survey (NMES)
(1987). The results from the benchmarking exercise are shown in Table 10.
In general, AHEAD93 did a good job of estimating prevalence as found in other
representative studies of older people. The prevalence rates for walking several blocks, bathing,
eating, walking across a room, using the toilet, preparing meals, and using the telephone as observed
in AHEAD93 were close to those observed in the SOA II and/or the NMES. The global measure of
disability status, which took into account the prevalence rates reported with the individual ADL and
IADL measures, had roughly the same distribution between the AHEAD93 and the SOA II. Note
that these prevalence rates were fairly consistent across studies despite differences in question
wording.
38
Table 10
Percent reporting any difficulty with ADLs and IADLs, among non-institutionalized persons: AHEAD93, SOA II
1995, and NMES 1987.
Measure
AHEAD93 Age 70+
SOA II 1995 Age 70+
NMES 1987 Age 70+
Nagi Measures Lifting 10 pounds Walking 2-3 blocks
33.1% 32.0%
18.1% 33.8%
-- --
ADLs Bathing Dressing Eating Walking Transferring Using the Toilet Percent with 1+ ADL difficulty
12.3% 13.0% 5.4% 23.3% 9.2% 4.7%
29.7%
13.7% 8.6% 2.5%
23.9% 11.4% 6.2%
28.4%
11.5% 6.1% 1.2% 9.5% 7.2% 4.4%
16.0%
IADLs Preparing meals Shopping Using the telephone Managing money Percent with 1+ IADL difficulty
9.4% 18.1% 5.2% 11.0%
22.9%
8.7% 14.2% 4.6% 5.7%
17.0%
9.5% 13.8% 5.7% 8.0%
--
Disability status: No difficulty IADL difficulty only ADL difficulty
64.2% 6.1% 29.7%
68.2% 3.4%
28.4%
-- -- --
Notes: AHEAD: Does anyone ever help you [do ADL]? Do you ever use equipment or devices such as [xx] when [doing ADL]? (Even when someone helps you) (Even when using the equipment) (Without any help or special equipment) Do you have any difficulty [doing ADL]? SOA II: Because of a health or physical problem, do you have any difficulty [doing ADL]? NMES: Do you have any difficulty [doing ADL] without help? Do you receive help from another person [to do ADL]? Do you use special equipment or aids to [do ADL]?
39
G. Constructed variables For convenience, it is often desirable to have constructed variables available in publicly
available data sets. Currently, HRS and AHEAD staff have constructed summary variables for ADLs
and IADLs in AHEAD93 only. This section briefly lists the names and purpose of those variables.
The technical appendix provides SAS code for reconstructing most of these variables in the
AHEAD93 data set (except for the flags, which we believe were redundant) and for constructing
comparable variables for HRS92-00 and AHEAD95.
The names for the constructed ADL variables are WALK, DRESS, BATHE, EAT, BED,
TOILET, ADLANY, and NUMADL. The first five variables indicate whether respondents had
difficulty or needed help with the tasks for which they are named. ADLANY flags respondents who
had one or more difficulties and/or needed any help and NUMADL indicates the number of tasks
with which respondents had difficulty or needed help (Range: 0-6).
The constructed IADL variables in AHEAD92 differentiate between people who needed help
or couldn’t do the activity for health reasons from people who could do the activity or didn’t do it for
reasons unrelated to health. The SAS names for these variables are MEAL, GROC, PHONE,
MEDICINE, and MONEY. Also, there are constructed measures flagging whether respondents
needed help/had difficulty with any of the five IADLs (IADLANY) and summarizing the number of
IADLs for which they needed help or had difficulty (Range: 0-5) (NUMIADL).
For more summary variables, consult also the RAND Data Documentation (St. Clair et al.,
2003) and Data Sets. RAND provides cleaned and processed summary physical functioning
variables for HRS and AHEAD through HRS00.
40
VII. Technicalities
A. Strategies for accommodating changes in questions There are several approaches that users might consider to deal with the challenges posed by
changes in the sequence and wording of the questions. One is to limit observations to respondents
who answered identical questions or over time. This would be possible by looking at respondents
who completed a module on physical functioning, such as the LSOA module (Module 3) at the end of
AHEAD93 and AHEAD95 and the AHEAD93 modules (Modules 1 and 2) at the end of AHEAD95
and HRS98. Similarly, analysts might use only those questions that are consistent (if not identical)
over time, such as those ascertaining difficulty. Reynolds and Silverstein (2003) used this method to
examine the onset of disability in ADL and IADL function among AHEAD respondents. They
examined change occurring between AHEAD93 and AHEAD95 and between AHEAD93 and
HRS98.
Another approach is to limit observation to changes that occurred between waves after the
question format stabilized. Fonda and colleagues (2002) adopted this strategy in a paper examining
patterns in physical functioning among HRS and AHEAD respondents. Specifically, they delineated
patterns of change between HRS96 and HRS98 and between AHEAD95 and HRS98.
Item response theory (IRT, also known as latent trait theory) and structural equation modeling
(particularly path analysis with latent variables) might provide another option for dealing with
changes in the questions across waves. Using a series of probabilistic models, IRT and structural
equation models relate question responses to the latent traits presumed to underlie such responses.
IRT separately estimates person parameters and item parameters, and this property can be used to
generate estimates of an underlying physical functioning trait that are on a similar metric, even if
different questionnaire formats are used across repeated administration. Structural equation modeling
techniques, assumes that the observed variables (e.g., individual ADLs, IADLs, and
41
mobility/strength/fine motor skills) are imperfect indicators (i.e., they contain measurement error) for
the latent construct, so slight variation in question wording should be relatively inconsequential.
Several papers demonstrate how IRT might be used to work with the physical functioning
measures in HRS and AHEAD. For example, McHorney (2002) examined whether three sets of
physical functioning questions used in AHEAD93 could be equated such that scores from the
different questions related on a common metric or underlying ability distribution. One set of
questions included those contained in the core of the survey asked of all survey participants, another
set was based on the LSOA module questions which were asked of about 10% of the survey
participants in addition to the core questions, and the third set was based on the NLTCS module
questions which were asked of about 10% of the survey participants in addition to the core questions.
From the core, McHorney included questions covering receipt of help in ADLs, difficulty with
“higher-order” ADLs, difficulty with and/or inability to perform IADLs, and use of accommodative
equipment for and difficulty with ADLs not considered “higher-order”. She used all questions from
the modules. The analyses involved assessing unidimensionality and eliminating items that formed
additional factors, comparising model fit of 1-parameter versus 2-parameter models, and examining
differential item functioning and subsequently calibrating the physical functioning measures. In
brief, McHorney found that the core and module questions could be linked and placed on a single,
underlying distribution. Although McHorney’s study investigated one wave of data, it has
implications for how one might deal with the changes in the question wording over time for
longitudinal analyses. First, given that later waves adopted a question format for the physical
functioning measures that is very similar to that in the LSOA, it is likely that HRS92 and AHEAD93
questions form metrics of physical functioning are comparable to the physical functioning metrics
formed in later waves (McHorney demonstrated that these different questions relate to the same
underlying distribution for 1 wave, or were “equivalent”). Second, although not all respondents
42
answered the same questions, the IRT analysis generated known item parameters which can be used
to estimate respondents’ latent trait level of difficulty in physical functioning. If such an analysis
were performed on later waves, equivalent latent trait “scores” can be used to examine change in
physical functioning over time.
An analysis of the depressive symptom measures in the HRS (Jones and Fonda, in press)
provides another example of how to use IRT to deal with changes in survey questions over time. The
HRS included items from the Centers for Epidemiologic Studies - Depression scale (CES-D) to
address depressive symptoms in this cohort. As with the physical functioning measures, CES-D
symptom coverage and response categories varied across study wave. Jones and Fonda applied a IRT
based structural equation model to generate linked depressive symptom scores and then used
generalized estimating equation models to characterize trajectories of linked depressive symptom
scores with age within strata defined by race/ethnicity, sex and birth year. A similar process might be
followed for the physical functioning measures.
Currently HRS and AHEAD staff are considering another approach for dealing with changes
in the questions about ADLs, which would provide users with constructed variables. This approach
would involve the imputation of “new” HRS92 and AHEAD93 ADL scores. It would use
information from the 1994 and 1995 cores of HRS and AHEAD (observed separately) as well as
information from the 1994 modules that repeated the question format used in HRS92/AHEAD93
(i.e., Module 3 of HRS94 and Modules 1 and 2 of AHEAD95). All of this information would be
incorporated, either as an outcome or predictor, into logistic regression models predicting whether
respondents had difficulty with tasks. To illustrate, there were fourteen ADL questions in the core of
AHEAD95; one difficulty question for each of the six ADLs, one help question for each ADL, and
two questions about the use of assistive devices (for walking and getting in/out of bed). These
fourteen, core ADL questions would correspond to the dependent variables in fourteen logistic
43
regression models. In these models, the predictors would be dichotomous variables based on the
answers to the ADL questions in Modules 1 and 2. HRS and would first predict to whether
respondents reported any difficulty with each ADL question, using only the module variables as
predictors. Then, when they predicted to whether or not respondents used equipment or help, they
would use the module questions and the core questions about difficulty. This process would be
repeated through the fourteen regression models for each AHEAD95 core ADL. The strengths and
limitations to this potential strategy for dealing with the change in questions over time have yet to be
explored fully, however.
B. SAS code This section contains SAS code for users to refer to as they make re-coding decisions about
the physical functioning measures in the HRS and AHEAD. As with other parts of this
documentation, this section is meant to guide, not prescribe, use of the data. Ultimately, decisions
about how to code and combine the data are the user’s responsibility. Further, since variables are
occasionally modified in the process of data cleaning, it is the responsibility of the user to check this
SAS code carefully for errors and consistency with the variable format they are planning.
******* HRS92 / WAVE 1 data temp1; set hrsw1.health; array funcw1{21} v304-v324; do i = 1 to 21; if funcw1(i) = 1 then funcw1(i) = 0; else if fincw1(i) > 1 then funcw1(i) = 1; end; *** Assign intuitive names; w1run = v304; w1sevblk = v305; w1block = v306; w1cross = v307; w1sit = v308;
44
w1getup = v309; w1bed = v310; w1stairs = v311; w1stair = v312; w1lift = v313; w1stoop = v314; w1dime = v315; w1bathe = v316; w1raise = v317; w1pull = v318; w1eat = v319; w1dress = v320; w1map = v321; w1micro = v322; w1calc = v323; w1comp = v324; *** Create summary measures; w1nagi = sum (of w1run w1sevblk w1block w1sit w1getup w1stairs w1stair w1lift w1stoop w1dime
w1raise w1pull) ; w1adl = sum (of w1cross w1bed w1bathe w1eat w1dress); w1iadl = sum (of w1map w1micro w1calc w1comp); run;
******* HRS94 / WAVE 2H; data temp2; set hrsw2.health; array func2 {22} w306-w327; do i = 1 to 22; if func2(i) in (0, 1, 2, 3, 4, 6) then func2(i) = 1; else if func2(i) = 5 then func2(i) = 0; else if func2(i) in (8, 9) then func2(i) = .; end; *** Assign intuitive names; w2run = w306; w2sevblk = w307; w2block = w308; w2cross = w309; w2sit = w310; w2getup = w311; w2bed = w312; w2stairs = w313; w2stair = w314; w2lift = w315;
45
w2stoop = w316; w2dime = w317; w2bathe = w318; w2raise = w319; w2pull = w320; w2eat = w321; w2dress = w322; w2map = w323; w2calc = w324; w2money = w325; w2phone = w326; w2meds = w327; *** Create summary variables; w2nagi = sum (of w2run w2sevblk w1block w2sit w2getup w2stairs w2stair w2lift w2stoop w2dime
w2raise w2pull); w2adl = sum (of w2cross w2bed w2bathe w2eat w2dress); w2iadl = sum (of w2map w2calc w2money w2phone w2meds); run;
*******HRS96 / WAVE 3H; data temp3; set hrsw3.health; *** Re-codes addressing skips in questions regarding mobility, strength, & fine motor skills; if q1858 = 1 then q1861 = 1; if q1858 = 5 the